Skip to content
Theme
OnDevice Tools

Tech BusinessChips & Semiconductors

Micron bets $10 billion on long-term memory research for the AI era

Micron is setting up Micron Research Labs in Boise with a planned $10 billion investment over ten years, aiming to push memory research beyond its usual five-to-seven-year roadmap.

Illustration of a chip research facility in Idaho with engineers examining memory wafer samples

Key takeaways

  • Micron will spend a planned $10 billion over ten years on Micron Research Labs in Boise, Idaho, with groundbreaking expected in 2027.
  • The lab will look beyond Micron's usual five-to-seven-year roadmap, covering future memory designs, memory-compute architectures, packaging and manufacturing.
  • Micron says AI applications from data centres to cars and robots are multiplying its research needs and driving closer work with key customers.
  • Analysts say rising node costs have reshaped the industry, and AI demand has pushed DRAM gross margins to about 80%.
In this article
  1. A new home for long-horizon research
  2. Why Micron is changing its approach
  3. AI raises the stakes
  4. The rising cost of staying in the game
  5. What it means for the industry

A new home for long-horizon research

Micron Technology has announced a dedicated research organisation that will look well beyond its usual product roadmap. Micron Research Labs, unveiled in the summer of 2026, will be based in Boise, Idaho, and is backed by a planned $10 billion investment spread over the next decade, according to EE Times.

The lab will work on the memory technologies that Micron sees as critical in the years ahead, along with new ways of combining memory and computing, advanced packaging methods, and future approaches to chipmaking. The spending plan also covers work with universities, smaller satellite sites and partnerships across the wider semiconductor ecosystem.

Micron expects to break ground on a building in 2027 that can house several hundred researchers. The company has already named Deirdre Hanford, who spent 37 years at the design software firm Synopsys, as corporate vice president and president of the new lab. Her job is to build up the research infrastructure, define how the lab operates and set up its partnerships.

Why Micron is changing its approach

R&D spending has always been heavy, but until now it mostly targeted challenges a few years out. He noted that the current level of investment would once have been hard to imagine.

According to DeBoer, Micron has always planned ahead, but mainly within roadmaps covering the next five to seven years. Work looking a decade or more into the future was left largely to others, partly because US government money flowed generously into university research. Now that memory has become far more important to the industry, Micron wants to lead that longer-term effort itself.

In recent years the company concentrated on supporting professors and students, which helped build the workforce it needs. The new lab goes further: DeBoer described it as a pivot towards research with concrete outcomes in mind, while avoiding some of the intellectual property tangles that come with projects close to commercialisation. He added that the lab's funding sits on top of investments Micron has already announced or forecast, which have grown every year for the past four years.

An early sign of that direction came this autumn, when Micron said it intends to put up to $50 million over five years into the US Department of Energy's Genesis Mission Graduate Fellowship Pilot Program and related work at national laboratories.

AI raises the stakes

Illustration: AI, privacy, car, robot and the labels Dram, Imec, Micron

Much of the memory industry's research has traditionally focused on near-term manufacturing gains: shrinking DRAM from one process node to the next and improving yields. AI makes the longer view more urgent while also complicating the work that is close to production.

DeBoer argued that memory is the engine behind AI, whether in data centres or at the edge in cars and robots. All of these applications depend heavily on memory performance, bandwidth and power use. The sheer range of uses multiplies the research burden. In data centres, the capacity and performance expectations for LPDDR, a low-power type of memory, have changed sharply, while robotics and cars demand advanced packaging and three-dimensional designs.

These projects, he said, involve close work with key customers rather than simply delivering another scaled-down DRAM node. In his view, a system cannot be optimised unless the memory inside it is optimised too.

The rising cost of staying in the game

Analysts interviewed by EE Times see the move as a rational response to industry economics. Dan Hutcheson, vice chair at research firm TechInsights, said the new lab is a sensible way to bring in outside funding and ideas. Shared research organisations such as Sematech and imec eased the problem, but the expense still pushed weaker vendors out of the market.

Hutcheson also drew a distinction between basic research and development work. He argued that companies rarely fund deep basic research because it can take 10 to 20 years to mature, which is why universities, national labs and organisations such as imec matter. In his view, imec has done well at bridging the gap between universities, suppliers and customers. He added that AI has turned memory from a cost-driven business into a value-driven one.

Jim Handy, principal analyst at Objective Analysis, offered a more sceptical reading. He noted that a DRAM maker typically puts about 20% of revenue into capital spending and about 10% into R&D, with few layoffs in the research team even in downturns. In his view, the $10 billion pledge is partly existing spending repackaged as a ten-year commitment.

What it means for the industry

Two cost centres shape memory research: developing the technology itself, and paying for the factories that produce it. Handy said consolidation among memory vendors is driven mainly by the rising cost of building a competitive fab, which has outpaced DRAM revenue growth.

Even so, the current market gives memory makers unusual room to invest. Demand from large AI players, who are less sensitive to price, is delivering gross margins of around 80%, far above the 30% or lower that DRAM makers have historically accepted, according to Handy. More research spending should also help manufacturers move to new nodes faster and extract more bits from existing capacity, which he expects will eventually lower prices for buyers.

For Micron, the lab signals that memory is now central to the AI build-out rather than a commodity add-on. Whether the bet pays off depends on turning decade-long research into products that data centre and device makers actually want.

Sources

  1. Memory R&D Is a High-Stakes GameEE Times

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.